Codex や deep research といったエージェントの例からもわかるように、リーズニングモデルが複雑な問題を解くには数分かかることがあります。バックグラウンドモードを使うと、タイムアウトなどの接続上の問題を気にせず、GPT-5.2 や GPT-5.2 Pro などのモデルで時間のかかるタスクを安定して実行できます。
バックグラウンドモードではタスクが非同期で開始されます。レスポンスオブジェクトをポーリングすると、タスクの状態を随時確認できます。バックグラウンドでレスポンスの生成を開始するには、background を true に設定して API リクエストを送信します。
ゼロデータ保持（ZDR）が適用されるプロジェクトでは、バックグラウンドリクエストは store=false で実行されます。非同期実行とポーリングを可能にするため、レスポンスデータは約 10 分間ディスクに一時保存されます。
変更された不正利用監視（強化版を含む）を使用するプロジェクトでは、フォアグラウンドリクエストで store を省略するか true に設定すると、標準のデータ保持ポリシーが適用されます。バックグラウンドレスポンスがポーリング期間の終了後も保持されるのは、store=true を明示的に指定した場合のみです。バックグラウンドリクエストで store を省略するか false に設定すると、レスポンスは約 10 分後に削除されます。
バックグラウンドでのレスポンス生成
Python
1
2
3
4
5
6
7
8curl https://api.openai.com/v1/responses \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-d '{
"model": "gpt-6-astra",
"input": "Write a very long novel about otters in space.",
"background": true
}'1
2
3
4
5
6
7
8
9
10import OpenAI from "openai";
const client = new OpenAI();
const resp = await client.responses.create({
model: "gpt-6-astra",
input: "Write a very long novel about otters in space.",
background: true,
});
console.log(resp.status);1
2
3
4
5
6
7
8
9
10
11from openai import OpenAI
client = OpenAI()
resp = client.responses.create(
model="gpt-6-astra",
input="Write a very long novel about otters in space.",
background=True,
)
print(resp.status)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
response, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-6-astra",
Background: openai.Bool(true),
Input: responses.ResponseNewParamsInputUnion{
OfString: openai.String("Write a very long novel about otters in space."),
},
})
if err != nil {
panic(err)
}
fmt.Println(response.Status)
}1
2
3
4
5
6
7
8
9
10
11
12
13import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.responses.ResponseCreateParams;
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6-astra")
.input("Write a detailed market analysis.")
.background(true)
.build();
var response = client.responses().create(params);
System.out.println(response.status().orElseThrow());1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
CreateResponseOptions options = new()
{
Model = "gpt-6-astra",
BackgroundModeEnabled = true,
};
options.InputItems.Add(
ResponseItem.CreateUserMessageItem("Write a very long novel about otters in space.")
);
ResponseResult response = await client.CreateResponseAsync(options);
Console.WriteLine(response.Status);1
2
3
4
5
6
7
8
9
10require "openai"
client = OpenAI::Client.new
response = client.responses.create(
model: "gpt-6-astra",
input: "Write a detailed market analysis.",
background: true
)
puts(response.status)バックグラウンドレスポンスのポーリング
バックグラウンドリクエストの状態を確認するには、Responses API の GET エンドポイントを使用します。リクエストの状態が queued または in_progress の間は、ポーリングを続けてください。これら以外の状態に移行した時点で、リクエストは終了状態に達しています。
バックグラウンドで生成中のレスポンスの取得
Python
1
2
3curl https://api.openai.com/v1/responses/resp_123 \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY"1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16import OpenAI from "openai";
const client = new OpenAI();
let resp = await client.responses.create({
model: "gpt-6-astra",
input: "Write a very long novel about otters in space.",
background: true,
});
while (resp.status === "queued" || resp.status === "in_progress") {
console.log("Current status: " + resp.status);
await new Promise((resolve) => setTimeout(resolve, 2000)); // wait 2 seconds
resp = await client.responses.retrieve(resp.id);
}
console.log("Final status: " + resp.status + "\nOutput:\n" + resp.output_text);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17from openai import OpenAI
from time import sleep
client = OpenAI()
resp = client.responses.create(
model="gpt-6-astra",
input="Write a very long novel about otters in space.",
background=True,
)
while resp.status in {"queued", "in_progress"}:
print(f"Current status: {resp.status}")
sleep(2)
resp = client.responses.retrieve(resp.id)
print(f"Final status: {resp.status}\nOutput:\n{resp.output_text}")1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36package main
import (
"context"
"fmt"
"time"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
response, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-6-astra",
Background: openai.Bool(true),
Input: responses.ResponseNewParamsInputUnion{
OfString: openai.String("Write a very long novel about otters in space."),
},
})
if err != nil {
panic(err)
}
for response.Status == "queued" || response.Status == "in_progress" {
fmt.Println("Current status:", response.Status)
time.Sleep(2 * time.Second)
response, err = client.Responses.Get(context.Background(), response.ID, responses.ResponseGetParams{})
if err != nil {
panic(err)
}
}
fmt.Printf("Final status: %s\nOutput:\n%s\n", response.Status, response.OutputText())
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseStatus;
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6-astra")
.input("Write a very long novel about otters in space.")
.background(true)
.build();
var response = client.responses().create(params);
while (response.status().filter(ResponseStatus.QUEUED::equals).isPresent()
|| response.status().filter(ResponseStatus.IN_PROGRESS::equals).isPresent()) {
System.out.println("Current status: " + response.status().orElseThrow());
Thread.sleep(1000);
response = client.responses().retrieve(response.id());
}
System.out.println("Final status: " + response.status().orElseThrow());
response.output().stream()
.flatMap(item -> item.message().stream())
.flatMap(message -> message.content().stream())
.flatMap(content -> content.outputText().stream())
.forEach(text -> System.out.println(text.text()));1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
CreateResponseOptions options = new()
{
Model = "gpt-6-astra",
BackgroundModeEnabled = true,
};
options.InputItems.Add(
ResponseItem.CreateUserMessageItem("Write a very long novel about otters in space.")
);
ResponseResult response = await client.CreateResponseAsync(options);
while (response.Status is ResponseStatus.Queued or ResponseStatus.InProgress)
{
await Task.Delay(TimeSpan.FromSeconds(1));
response = await client.GetResponseAsync(response.Id);
}
if (response.Status != ResponseStatus.Completed)
{
throw new InvalidOperationException($"Background response ended with status: {response.Status}");
}
Console.WriteLine($"Status: {response.Status}");
Console.WriteLine(response.GetOutputText());1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18require "openai"
client = OpenAI::Client.new
response = client.responses.create(
model: "gpt-6-astra",
input: "Write a very long novel about otters in space.",
background: true
)
while [:queued, :in_progress].include?(response.status)
puts("Current status: #{response.status}")
sleep(2)
response = client.responses.retrieve(response.id)
end
puts("Final status: #{response.status}")
puts("Output:")
puts(response.output_text)バックグラウンドレスポンスのキャンセル
次の例のように、生成中のレスポンスをキャンセルすることもできます。
生成中のレスポンスのキャンセル
Python
1
2
3curl -X POST https://api.openai.com/v1/responses/resp_123/cancel \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY"1
2
3
4
5
6import OpenAI from "openai";
const client = new OpenAI();
const resp = await client.responses.cancel("resp_123");
console.log(resp.status);1
2
3
4
5
6
7
8
9
10import os
from openai import OpenAI
response_id = os.environ["OPENAI_RESPONSE_ID"]
client = OpenAI()
resp = client.responses.cancel(response_id)
print(resp.status)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
)
func main() {
client := openai.NewClient()
canceled, err := client.Responses.Cancel(context.Background(), "resp_123")
if err != nil {
panic(err)
}
fmt.Println(canceled.Status)
}1
2
3
4
5
6
7
8import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
String responseId = "resp_123";
var response = client.responses().cancel(responseId);
System.out.println(response.status());1
2
3
4
5
6
7
8
9
10
11using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
// Replace this illustrative ID with the background response to cancel.
string responseId = "resp_123";
ResponseResult response = await client.CancelResponseAsync(responseId);
Console.WriteLine(response.Status);1
2
3
4
5require "openai"
client = OpenAI::Client.new
response = client.responses.cancel("resp_123")
puts(response.status)キャンセル操作は冪等です。繰り返し呼び出しても、2 回目以降は最終状態の Response オブジェクトが返されるだけです。
バックグラウンドレスポンスのストリーミング
バックグラウンドで Response を作成し、すぐにイベントのストリーミングを開始できます。クライアント側でストリームが途切れる可能性があり、後から再開できるようにしておきたい場合に便利です。ストリーミングを開始するには、background と stream を両方とも true に設定して Response を作成します。後から再開できるように、各ストリーミングイベントで受け取る sequence_number に対応する「カーソル」を記録しておいてください。
現在、バックグラウンドレスポンスでは、同期レスポンスに比べて最初のトークンを受け取るまでに時間がかかります。今後数週間でこの差を縮められるよう、改善を進めています。
バックグラウンドレスポンスの生成とストリーミング
curl
1
2
3
4
5
6
7
8
9
10
11
12
13
14curl https://api.openai.com/v1/responses \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-d '{
"model": "gpt-6-astra",
"input": "Write a very long novel about otters in space.",
"background": true,
"stream": true
}'
// To resume:
curl "https://api.openai.com/v1/responses/resp_123?stream=true&starting_after=42" \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY"1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25import OpenAI from "openai";
const client = new OpenAI();
const stream = await client.responses.create({
model: "gpt-6-astra",
input: "Write a very long novel about otters in space.",
background: true,
stream: true,
});
let cursor = null;
let responseId = null;
for await (const event of stream) {
console.log(event);
// Keepalives use connection counters, not stored-event resume positions.
if (String(event.type) !== "keepalive") cursor = event.sequence_number;
if ("response" in event && event.response?.id) responseId = event.response.id;
}
// If the connection drops, resume streaming from the last stored-event cursor:
// const resumedStream = await client.responses.retrieve(responseId, {
// stream: true,
// starting_after: cursor,
// });
// for await (const event of resumedStream) { ... }1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28from openai import OpenAI
client = OpenAI()
# Fire off an async response but also start streaming immediately
stream = client.responses.create(
model="gpt-6-astra",
input="Write a very long novel about otters in space.",
background=True,
stream=True,
)
cursor = None
response_id = None
for event in stream:
print(event)
if event.type == "response.created":
response_id = event.response.id
# Keepalives use connection counters, not stored-event resume positions.
if event.type != "keepalive":
cursor = event.sequence_number
# If your connection drops, the response continues running and you can reconnect:
# with client.responses.stream(response_id=response_id, starting_after=cursor) as resumed:
# for event in resumed:
# print(event)
if event.type == "response.created":
response_id = event.response.id1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
stream := client.Responses.NewStreaming(context.Background(), responses.ResponseNewParams{
Model: "gpt-6-astra",
Background: openai.Bool(true),
Input: responses.ResponseNewParamsInputUnion{
OfString: openai.String("Write a very long novel about otters in space."),
},
})
defer stream.Close()
var cursor int64
var responseID string
for stream.Next() {
event := stream.Current()
fmt.Println(event.Type)
// Keepalives use connection counters, not stored-event resume positions.
if event.Type != "keepalive" {
cursor = event.SequenceNumber
}
if event.Response.ID != "" {
responseID = event.Response.ID
}
}
if err := stream.Err(); err != nil {
panic(err)
}
fmt.Printf("response %s last cursor %d\n", responseID, cursor)
// If the connection drops, resume streaming from the last cursor:
// resumed := client.Responses.GetStreaming(
// context.Background(),
// responseID,
// responses.ResponseGetParams{StartingAfter: openai.Int(cursor)},
// )
// for resumed.Next() {
// fmt.Println(resumed.Current().Type)
// }
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106import com.fasterxml.jackson.databind.json.JsonMapper;
import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.RequestOptions;
import com.openai.core.http.StreamResponse;
import com.openai.errors.OpenAIIoException;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseRetrieveParams;
import com.openai.models.responses.ResponseStreamEvent;
import java.time.Duration;
import java.util.concurrent.atomic.AtomicBoolean;
import java.util.concurrent.atomic.AtomicLong;
import java.util.concurrent.atomic.AtomicReference;
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6-astra")
.input("Write a very long novel about otters in space.")
.background(true)
.build();
AtomicLong lastSequenceNumber = new AtomicLong(-1);
AtomicReference<String> responseId = new AtomicReference<>("");
AtomicBoolean streamCompleted = new AtomicBoolean(false);
JsonMapper json = new JsonMapper();
// This long-running response can exceed the SDK's default request deadline.
RequestOptions streamOptions = RequestOptions.builder().timeout(Duration.ofMinutes(30)).build();
try (StreamResponse<ResponseStreamEvent> stream =
client.responses().createStreaming(params, streamOptions)) {
stream.stream()
.forEach(
event -> {
if (event.failed().isPresent()
|| event.incomplete().isPresent()
|| event.error().isPresent()) {
throw new IllegalStateException("Background response did not complete");
}
var data = json.valueToTree(event);
// Keepalives count connection traffic, not stored response events.
if (data.path("type").asText().startsWith("response.")
&& data.path("sequence_number").isIntegralNumber()) {
lastSequenceNumber.set(data.path("sequence_number").asLong());
}
event
.created()
.ifPresent(
created -> {
responseId.set(created.response().id());
System.out.println("response.created");
});
event
.outputTextDelta()
.ifPresent(
delta -> {
System.out.println("response.output_text.delta");
});
event
.completed()
.ifPresent(
completed -> {
streamCompleted.set(true);
System.out.println("response.completed");
});
});
} catch (OpenAIIoException error) {
// Resume an interrupted stream once, from its stored-event checkpoint.
if (Thread.currentThread().isInterrupted()
|| streamCompleted.get()
|| responseId.get().isBlank()
|| lastSequenceNumber.get() < 0) {
throw error;
}
}
System.out.println(
"Response " + responseId.get() + "; last sequence number " + lastSequenceNumber.get());
if (!streamCompleted.get()) {
if (Thread.currentThread().isInterrupted()
|| responseId.get().isBlank()
|| lastSequenceNumber.get() < 0) {
throw new IllegalStateException("Stream ended without a resumable checkpoint");
}
try (StreamResponse<ResponseStreamEvent> resumed =
client
.responses()
.retrieveStreaming(
ResponseRetrieveParams.builder()
.responseId(responseId.get())
.startingAfter(lastSequenceNumber.get())
.build(),
streamOptions)) {
resumed.stream()
.forEach(
event -> {
if (event.failed().isPresent()
|| event.incomplete().isPresent()
|| event.error().isPresent()) {
throw new IllegalStateException("Resumed response did not complete");
}
event.outputTextDelta().ifPresent(delta -> System.out.println(delta.delta()));
event.completed().ifPresent(completed -> streamCompleted.set(true));
});
}
if (!streamCompleted.get()) {
throw new IllegalStateException("Resumed response did not complete");
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
CreateResponseOptions options = new()
{
Model = "gpt-6-astra",
BackgroundModeEnabled = true,
StreamingEnabled = true,
};
options.InputItems.Add(
ResponseItem.CreateUserMessageItem("Write a very long novel about otters in space.")
);
string? responseId = null;
int lastSequenceNumber = -1;
bool completed = false;
void HandleUpdate(StreamingResponseUpdate update)
{
lastSequenceNumber = update.SequenceNumber;
switch (update)
{
case StreamingResponseCreatedUpdate created:
responseId = created.Response.Id;
break;
case StreamingResponseOutputTextDeltaUpdate text:
Console.Write(text.Delta);
break;
case StreamingResponseCompletedUpdate:
completed = true;
break;
case StreamingResponseFailedUpdate:
throw new InvalidOperationException("The background response failed.");
case StreamingResponseIncompleteUpdate:
throw new InvalidOperationException("The background response was incomplete.");
case StreamingResponseErrorUpdate error:
throw new InvalidOperationException($"The response stream failed: {error.Message}");
}
}
try
{
await foreach (
StreamingResponseUpdate update in client.CreateResponseStreamingAsync(options)
)
{
HandleUpdate(update);
}
}
catch (Exception error)
when (error is HttpRequestException or IOException && responseId is not null)
{
// The background response continues after its streaming connection is interrupted.
}
if (!completed)
{
if (responseId is null)
{
throw new InvalidOperationException("The response stream ended before providing its ID.");
}
GetResponseOptions resumeOptions = new(responseId)
{
StartingAfter = lastSequenceNumber,
StreamingEnabled = true,
};
await foreach (StreamingResponseUpdate update in client.GetResponseStreamingAsync(resumeOptions))
{
HandleUpdate(update);
}
if (!completed)
{
throw new InvalidOperationException(
"The resumed response stream ended before the background response completed."
);
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31require "openai"
# An uncapped background novel can run beyond the SDK's default ten minutes.
client = OpenAI::Client.new(timeout: 30 * 60)
stream = client.responses.stream(
model: "gpt-6-astra",
input: "Write a very long novel about otters in space.",
background: true
)
last_sequence_number = -1
response_id = ""
begin
stream.each do |event|
puts(event.type)
# The SDK filters transport keepalives before yielding stored events.
last_sequence_number = event.sequence_number || last_sequence_number
if event.is_a?(OpenAI::Models::Responses::ResponseCreatedEvent)
response_id = event.response.id
end
end
ensure
stream.close
end
puts("response #{response_id} last cursor #{last_sequence_number}")
# If the connection drops, resume from the last sequence number:
# client.responses.stream(response_id: response_id, starting_after: last_sequence_number).each do |event|
# puts(event.type)
# end制限事項
- バックグラウンドリクエストでは
store=falseを指定できますが、非同期実行とポーリングのためにレスポンスデータは一時保存されます。 - 同期レスポンスをキャンセルするには、接続を切断してください。
- バックグラウンドレスポンスから新しいストリームを開始できるのは、レスポンスの作成時に
stream=trueを指定した場合のみです。